Building the loan origination platform thousands of partners ran on
B2B2C loan origination platform · API & partner integrations
A loan origination platform's growth was capped by how long it took to onboard a new partner — weeks of manual integration work per partner, which didn't scale against an ambitious distribution plan.
The fix was a configurable, API-led onboarding layer instead of bespoke integration work each time: a documented API suite, standard onboarding flows, and a payments and reconciliation layer that worked the same way across every gateway.
Launching a digital Gold Loan 0-to-1 inside a large consumer app
Secured lending · 0-to-1 launch & signal-driven targeting
There was a large eligible base sitting inside the app, a very large existing user base, but no secured-lending product to serve them, and no playbook for turning a cold, eligible user into a funded gold loan. The build was genuinely 0-to-1: the end-to-end digital journey, the integration with the lending partner, and the decisioning that decided who to approach and how.
Rather than push the same offer to everyone, I designed the targeting and onboarding logic — which signals to weigh, and in what order — using credit-line indicators, bureau data, live location, network-level signals, and moments-based triggers. The modelling sat with data science; the product work was shaping the workflow and the rules around it, then tightening the journey wherever users were dropping.
Matching the right partner to the right relationship manager
B2B2C lending operations · Signal-based matching
Across tens of thousands of sub-DSA partners spread over thousands of pincodes, routing partners to relationship managers by hand didn't scale — and bad matches quietly cost conversion further down the funnel, between a partner accepting terms and the loan actually disbursing.
I designed a matching engine that paired each DSA to an RM using pincode, credit eligibility, the RM's coverage area, and their current bandwidth — defining the matching rules and the logic for trade-offs when signals conflicted, then partnering with data science to run it at scale. Practitioner decisioning, not model-building.
Getting co-branded cards activated, not just issued
Co-branded EMI card · Activation & AI-led recommendations
Issuing a card is the easy half. The harder half is getting it activated and used — closing the gap between cardholders who simply hold the card and the smaller set actually transacting on consumer-durable loans. Issuance had scaled into the hundreds of thousands of cards cumulatively, with strong monthly peaks, so the constraint had clearly moved from acquisition to activation.
The lever was relevance: surfacing the right store deals and OEM offers to each cardholder based on their credit profile, location, and network signals. I owned the decisioning design — which offers go to whom, and when — working with data science to operationalise it, alongside in-store activation at partner counters. Activation was tracked by cohort rather than as a single blended number.
Risk-based onboarding: which checks, for which user
Co-branded EMI card · Onboarding & KYC
Onboarding drop-off was concentrated in the verification steps, and the obvious read was a UX problem. Cutting the data by segment told a different story: new-to-credit and thin-file users were failing for different reasons, at different points — a one-size-fits-all KYC flow mismatched against customers with very different risk profiles.
Risk and the lending partner wanted more checks; growth wanted fewer. Both were right. The funnel was instrumented to quantify what each incremental check cost in activation versus caught in fraud, and that evidence aligned everyone around a risk-based step-up flow — lighter verification where risk was low, stricter where it spiked.
Halving partner query volume with product-led fixes
B2B2C loan origination · Partner self-serve
On the origination platform thousands of DSAs ran on, partners were raising more than ten queries per disbursed loan, where their application was, why it was declined, when commission would land. Every query was a product failure, and ops was drowning in them.
I treated contact ratio as a product metric, not an ops one. A live application tracker with plain-language statuses, a self-serve document-correction flow, and a real-time payout dashboard removed the reasons partners were calling in the first place.
Taking a digital card product offline, into a nationwide store network
Phygital distribution · POS & in-store activation
A card product that worked well digitally had a ceiling: customers who'd rather discover and activate it in person, at a physical counter, with store staff guiding them. That meant designing for a completely different environment — POS integration, QR-based in-store onboarding, and merchant activation workflows, not just porting the app screen to a counter.
The result diversified acquisition beyond digital-only channels and brought down customer acquisition cost, while proving the same credit and KYC logic could hold up in a messier, offline-first environment.
Cutting order-related customer queries in half
D2C e-commerce · Post-purchase self-serve
On a fast-scaling D2C platform, the order-to-contact ratio was climbing as orders grew. Most tickets were the same handful of questions: where is my order, how do I return this, when will my refund arrive.
Rather than grow support headcount in line with order volume, I closed the gaps at the source: proactive status updates at each milestone, self-serve tracking and returns, and refund visibility. Self-serve scales with volume; headcount doesn't.